T. Johnson and D. Shasha. B-trees with inserts and deletes: Why free-at-empty is better than merge-at-half. Journal of Computer and System Sciences, 47:45--76, August 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....are compulsorily re inserted into the tree level to which they belong, resulting in O(log nb) performance in the worst case. It is important to note that B trees follow a similar behavior in industrial applications, i.e. they do not perform node mergings but only free nodes when they are empty [17]. 4. DEFINITION, UPDATE AND QUERY OPERATIONS OF LR TREES In this section we provide the formal definition of LRtrees, as well as the algorithms for the update and the query operations. Typically, an LR tree is a collection of a varying, yet bounded from above, number of wR trees. DEFINITION 3. ....

Johnson, T. and Shasha, D. (1993) B-trees with inserts and deletes: why free-at-empty is better than merge-at-half. J. Comput. Syst. Sci., 47, 45--76.

....we study scenarios in which I A is clustered (i.e. table R is sorted according to attribute A) and situations in which I A is not clustered. Typically the height of all indices is three and the height of the indices does not change even if 20 of the records are deleted. As proposed in [9], we only reorganize and garbage collect an index page if it is totally empty (We do not apply the techniques proposed in Section 2.3) In all experiments, almost no reorganization is carried out because of the random distribution of the keys of the records that are deleted. 4.2. Experiment 1: ....

T. Johnson and D. Shasha. B-trees with inserts and deletes: Why free-at-empty is better than merge-at-half. Journal of Computer and System Sciences, 47(1):45--76, Aug. 1993.

....that we discuss are restoration of clustering, purging of old data, creation of a backup copy, compaction, and construction of indexes. A technical report [12] surveys IBM and non IBM work in online reorganization. 2 Restoration of Clustering for DB2 for MVS ESA Database 2 (DB2) for MVS ESA [2] is a relational DBMS. The sections below describe several methods of reorganization for DB2, roughly in increasing order of concurrency. 2.1 Clustering and Reorganization In DB2, each table can have a clustering index. DB2 tries to assign data records to pages to reflect the data records order ....

....people discussed issues in reorganization and or reviewed parts of an earlier draft of this paper. References [1] Bennett, B. T. and Franaszek, P. A. Permutation Clustering: An Approach to On Line Storage Reorganization, IBM J. Research and Development, Vol. 21, No. 6, Nov. 1977, pp. 528 533. [2] Haderle, D. J. and Jackson, R. D. IBM Database 2 Overview, IBM Syst. J. Vol. 23, No. 2, 1984, pp. 112 125. 3] IBM Corp. An Introduction to DataPropagator Relational Release 1, GC26 3398 01, 1993. 4] IBM Corp. Implementing Concurrent Copy, GG24 3990 00, Dec. 1993. 5] IBM Corp. and Integrated ....

[Article contains additional citation context not shown here]

Theodore Johnson and Dennis Shasha. B-trees with inserts and deletes: Why free-at-empty is better than merge-at-half. Journal of Computer and System Sciences, 47(1):45--76, 1993.

....the B tree is sparsely populated, that is, a large portion of many leaf pages is unused. There are also some free pages available in the database, which are not connected to the B tree. We assume that there is no ongoing node consolidation in the B tree, that is, the free at empty [JS93] policy is adapted for leaf pages. In the free at empty policy, non empty sparse nodes are never consolidated, but when a node becomes completely empty, its page is deallocated and its parent is updated to reflect this. Most commercial systems adapt this policy. Since availability of the database ....

Theodore Johnson and Dennis Shasha. B-trees with inserts and deletes: Why free-at-empty is better than merge-at-half. Journal of Computer and System Sciences, 47(1):45--76, 1993.

....the B tree is sparsely populated, that is, a large portion of many leaf pages is unused. There are also some free pages available in the database, which are not connected to the B tree. We assume that there is no ongoing node consolidation in the B tree, that is, the free at empty [JS93] policy is adapted for leaf pages. In the free at empty policy, nonempty sparse nodes are never consolidated, but when a node becomes completely empty, its page is deallocated and its parent is updated to reflect this. Most commercial systems adapt this policy. Since availability of the database ....

Theodore Johnson and Dennis Shasha. B-trees with inserts and deletes: Why free-at-empty is better than merge-at-half. Journal of Computer and System Sciences, 47(1):45--76, 1993.

....but there is no possibility of placing the record on some other existing leaf page. While this organization makes space allocation straight forward, it does not lead to particularly good space utilization: Johnson and Shasha show typical space utilization for B trees to be between 39 and 70 [JS93] In a B tree file organization, records do not have fixed physical addresses; they may be moved from page to page as other records are inserted into or deleted from the relation. The lack of fixed addresses presents a problem in the design of secondary indices, which need to point to ....

T. Johnson and D. Shasha. B-Trees with Inserts and Deletes: Why Free-at-Empty Is Better Than Merge-at-Half. Journal of Computer and System Sciences, 47:45--76, 1993.

....research which considered insertions only [BY89] The simulations in [JS89] support its analysis to show that in typical situations, where deletions don t outnumber insertions in the mix of operations, the tree nodes will contain acceptable percentages of entries. One of the work s assumptions [JS93a] is that the keys and tree operations are chosen uniformly from a random distribution. This assumption is unreasonable in certain realistic situations such as one described below. Allowing interior nodes with only a single pointer to exist in a B tree creates the possibility for arbitrarily ....

T. Johnson and D. Shasha. B-trees with inserts and deletes: Why free-at-empty is better than merge-at-half. Journal of Computer and System Sciences, 47(1):45--76, 1993.

No context found.

T. Johnson and D. Shasha. B-trees with inserts and deletes: Why free-at-empty is better than merge-at-half. Journal of Computer and System Sciences, 47:45--76, August 1993.

No context found.

T. Johnson, D. Shasha. B-Trees with Inserts and Deletes: Why-Free-at-Empty is better than Merge-at-Half Journal of Computer and System Sciences, 47: pp. 45-76, 1993.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC